Known for the methods of recommending television programs and radio programs are initial interest registering, viewing log using, and emphasis filtering, for example.
In each of these methods, the source data is EPG (Electronic Program Guide) information or program information (program metadata) on the Web for example. The methods are classified into the above-mentioned three methods depending on how the user's preference data to be matched with these pieces of information are obtained.
In the initial interest registering method, user's favorite categories (such as drama and variety for example), favorite genre names (such as drama and music for example), and favorite entertainers' names for example are registered by the user at the time of starting the use of a recommendation service. Subsequently, matching is executed with the program metadata by use of the registered information as keywords, thereby acquiring program names to be recommended.
In the viewing log using method, every time the user views programs, the program metadata about each viewed program are accumulated, and when a predetermined amount of viewing log (or program metadata) are accumulated, the accumulated viewing log is analyzed to acquire program names for recommendation. With a device on which video recording is made onto its hard disk drive for example, an operation log such as timer video recording and starting of video recording for example by the user may be used instead of the above-mentioned viewing log. In this case, the information highly reflecting user's interest can be obtained, rather than vague program information.
In the emphasis filtering method, the viewing (or operation) log of one user is matched with the viewing logs of other users to acquire viewing logs of other users which are similar to the viewing log of the user concerned. Then, of the programs viewed by other users similar in viewing log (namely, similar in preference) to the viewing of the user concerned, those program names which have not been viewed by the user concerned are obtained for recommendation.
Use of the above-mentioned known program recommendation methods allows the recommendation of programs in which each user seems to be interested.
However, each of the above-mentioned known recommendation methods comes to extract user's interest from program metadata (namely, resulting in the acquisition of lopsided interests in television programs). And, in the structure of program metadata, each of these methods uses generally intelligible program names, thereby presenting a problem that similarly sounding programs are recommended.
Namely, each of the above-mentioned known recommendation methods cannot reflect user's daily interests to the programs, thereby failing to recommend timely and useful programs.
At the same time, each of these methods presents a problem that, when particular programs are recommended, the user cannot understand the reason of the recommendation.